Paper Tribe or Nation Nation Building and Public Goods in Kenya Versus Tanzania

TRIBE OR NATION?
Nation Building and Public Goods in

Kenya versus Tanzania
By EDWARD MIGUEL*
I. INTRODUCTION
T HE design of public policies that promote interethnic cooperation

remains poorly understood nearly twenty years after the appear-
ance of Horowitz's seminal work.1 Recent research suggests that ethni-
cally diverse societies are prone to corruption, political instability, poor
institutional performance, and slow economic growth and that in the
United States higher levels of diversity are related to lower provision of
local public goods across municipalities. Addressing ethnic divisions is
likely to be particularly important for Africa, the most ethnically di-
verse and poorest continent.
This article examines how central government nation-building poli-
cies affect interethnic cooperation, by comparing the relationship be-
tween local ethnic diversity and public goods across two nearby rural
districts, one in western Kenya and one in western Tanzania, using
colonial-era national boundary placement as a "natural experiment."
Despite their largely shared geography, history, and colonial institu-
tional legacy, governments in Kenya and Tanzania have followed radi-
cally different ethnic policies along a range of dimensions—most
notably in national language policy, the educational curriculum, and
local institutional reform—with Tanzania consistently pursuing the
more serious nation-building policies during the postcolonial period.
The empirical evidence in this article suggests that the Tanzanian
nation-building approach has allowed ethnically diverse communities
'George Akerlof, Abhijit Banerjee, Robin Burgess, Tina Green, David Karol, David Laitin, Dan
Posner, Gerard Roland, Shanker Satyanath, and anonymous referees have provided valuable com-
ments. I am deeply indebted to the staff of ICS Africa in both Kenya and Tanzania, the Meatu District
Council, the Dutch Rural Development Programme, Alicia Bannon, Elizabeth Beasley, James Hab-
yarimana, Sylvie Moulin, Avery Ouellette, Polycarp Waswa, and especially Melissa Gonzalez-Brenes,
Mary Kay Gugerty, and Michael Kremer. Tina Green and Negar Ghobadi provided excellent research
assistance. I am grateful for support from the National Science Foundation (SGER-#0213652) and
the U.C. Berkeley Committee on Research. All errors remain my own.
'Donald Horowitz, Ethnic Groups in Conflict (Berkeley: University of California Press, 1985).
World Politics 56 (April 2004), 327-62

TRIBE OR NATION? 329
tion, slow financial development, poor schooling outcomes, and less in-
frastructure investment.
Empirical researchers have since documented many specific in-
stances where ethnic diversity produces subpar collective action out-
comes, and for the remainder of this section I survey this growing body
of evidence. To illustrate, Peruvian microcredit groups have higher loan
default rates when members are from different cultural backgrounds.4
U.S. municipalities with higher levels of racial diversity raise consider-
ably less funding for local public goods.5 Rural Kenyan communities
with greater ethnolinguistic diversity—across "tribes," as they are called
in East Africa—have considerably less primary school funding, worse
school facilities, and poor maintenance of water wells.6
There is less consensus regarding the underlying mechanisms gener-
ating these patterns, and two sets of theories have emerged, although
the theories are not mutually exclusive and both in most cases probably
capture important aspects of reality. The first theories are what I call
taste explanations for negative ethnic diversity effects. There are several
variants of this theory, mainly developed in research on the United
States. For example, Alesina, Baqir, and Easterly7 have argued that in-
dividuals from different ethnic groups prefer distinct types of public
goods—roads versus libraries, for instance—and this leads to less agree-
ment on public goods choices and thus to lower funding in diverse
areas; Alesina and La Ferrara8 have asserted that, for the most part, in-
dividuals from different groups dislike "mixing" across ethnic lines, and
this drives the poor collective outcomes in diverse areas; and Vigdor9
finds that individuals prefer to fund public goods that benefit their own
ethnic group. Unfortunately, none of these theories explain where eth-
nic taste differences come from or how they are affected by policy. They
therefore cannot address the central concern of this study—ameliorat-
ing ethnic divisions.
The second set of theories emphasizes the important role of com-
munity social sanctions in sustaining collective action and how diverse
4
Dean Karlan, "Social Capital and Group Banking" (Manuscript, Department of Economics,
Princeton University, 2002).
5
Alberto Alesina, Reza Baqir, and William Easterly, "Public Goods and Ethnic Divisions," Quar-
terly Journal ofEconomics 114, no. 4 (1999).
6
Edward Miguel and Mary Kay Gugerty, "Ethnic Diversity, Social Sanctions, and Public Goods in
Kenya" Journal ofPublic Economics (forthcoming).
7
Alesina, Baqir, and Easterly (fn. 5).
8
Alberto Alesina and Eliana La Ferrara, "Participation in Heterogeneous Communities," Quarterly
Journal ofEconomics 115, no. 3 (2000).
'Jacob Vigdor, "Community Composition and Collective Action: Analyzing Initial Mail Response
to the 2000 Census," Review ofEconomics and Statistics (forthcoming).
330 WORLD POLITICS
settings can render sanctions ineffective. Observers of less developed
countries have long noted the importance of community pressure and
dense social ties in sustaining good collective outcomes, and recent em-
pirical studies tend to emphasize this mechanism. The basic idea is that
it becomes difficult to sustain cooperation across ethnic groups in areas
where members of different groups tend not to have frequent social in-
teractions or personal affinity.10 In this view, public policies that promote
interaction, information sharing, and coordination across groups are
plausible vehicles for reducing the inefficiencies associated with diversity.
Yet there is limited empirical evidence regarding which public poli-
cies are most successful in addressing ethnic divisions. One possible in-
stitutional reform would be to promote power sharing across groups
within governments or other organizations. Under power sharing, eth-
nic minorities are assured some minimum representation in govern-
ment and some influence over policy, including veto power over certain
policies.11 Although intuitively attractive, power sharing has failed to
resolve ethnic conflict in many cases, especially in Africa.12 In fact, at
the same time that power sharing structures competition among ethnic
groups in the political arena, it institutionalizes divisions across groups
rather than bridging them and may hinder the development of new so-
cial identities—or multiethnic political coalitions—that cut across pre-
existing divisions.
A second and perhaps more promising approach advocates promot-
ing dialogue and interaction among the leaders of distinct ethnic com-
munities, who are then better able to coordinate responses to violations
of intergroup cooperation norms. One variant of this approach, associ-
ated with the work of Fearon and Laitin,13 would have group leaders
punish violators from within their own ethnic group, so-called within-
group policing. A closely related form of elite coordination is the estab-
lishment of formal associational bonds across ethnic groups. It has recently
been argued that the density of cross-group associational ties is the
critical determinant of interreligious relations in India during episodes
of communal violence.14
However, these theories too have limitations. Approaches predicated
on the existence of cooperation among ethnic group leaders beg the
10
Miguel and Gugerty (fn. 6).
11
Arend Lijphart, "Consociational Democracy," World Politics 21 (January 1969).
12
Ian Spears, "Africa: The Limits of Power-Sharing," Journal ofDemocracy 13, no. 3 (2002).
"James Fearon and David Laitin, "Explaining Interethnic Cooperation," American Political Science
Review 90, no. 4 (1996).
"Ashutosh Varshney, Civic Life and Ethnic Conflict: Hindus and Muslims in India (New Haven: Yale
University Press, 2002).
question of how this cooperation comes about in the first place. In fact,
elite cooperation is as much a manifestation of better ethnic relations
as a cause, and thus it is difficult to draw strong causal claims about
how associational links actually affect relations.
NATION BUILDING AND POLITICAL SOCIALIZATION
The literatures on nation building and political socialization have con-
cerned themselves with these and related questions, namely, with how
individual political ideals, opinions, identities, and preferences are cre-
ated. Research on political socialization has focused primarily on how
the mass media and the educational system can be employed by politi-
cal leaders to inculcate citizens with "desirable" political ideals, includ-
ing often a strong attachment to the nation over ethnic and regional
identities. Thus, to the extent that it is successful, political socialization
offers a way out of the dilemma of determining the origin of "tastes" for
ethnic cooperation. Nation-building policies may therefore be viewed
more generally as investments in "social capital."15
Nation-building reforms in the newly independent East African na-
tions figured prominently in the political socialization literature of the
1960s and 1970s.16 However, this body of research did not reach strong
empirical conclusions about how useful political socialization actually
was in manufacturing a coherent national identity or political culture,17
in part because of the limited time that had elapsed between the im-
plementation of nation-building programs in the 1960s and the re-
search conducted only a decade later.
III. A NATION-BUILDING CASE STUDY: KENYA VERSUS TANZANIA
GEOGRAPHIC AND HISTORICAL COMMONALITIES

Kenya and Tanzania are a natural paired comparison, with similar ge-
ography and histories, but they have followed radically different nation-
building policies since independence. Barkan writes:
Comparison between Kenya and Tanzania [is] ... appealing because of their re-
semblances with respect to a number of variables that impinge upon the devel-
opmental process and that could be held constant or nearly constant in an
13
Robert Putnam, Making Democracy Work (Princeton: Princeton University Press, 1993).
16
Refer to Kenneth G. Prewitt, George Von der Muhll, and David Court, "School Experiences and
Political Socialization: A Study ofTanzanian Secondary School Students," Comparative Political Stud-
ies 3, no. 2 (1970).
'' David Court and Kabiru Kinyanjui, "Development Policy and Educational Opportunity: The Ex-
perience of Kenya and Tanzania," IDS Occasional Paper, no. 33 (Nairobi: University of Nairobi, 1980).
332 WORLD POLITICS
Mombasa
UNITED REP. OF
TANZANIA
, Dares Salaam
32° 40-
FIGURE 1
M A P OF EAST AFRICA
(FEATURING BUSIA DISTRICT, KENYA, AND MEATU DISTRICT, TANZANIA)
examination of the countries. Both are populated mainly by small peasant

households of similar cultures Both experienced British colonial rule and in-
herited a common set of political, administrative, and economic institutions....
As adjacent countries, they share a common climate and have similar natural re-
source endowments.18
The two districts in this study—Busia, Kenya, and Meatu, Tanzania

(see Figure 1)—are the sites of field offices for the same Dutch non-
governmental organization (ICS Africa) and were originally chosen for
development assistance according to the same criteria, as poor rural
areas in particular need of assistance. Although the Tanzanian district is
18
Joel D. Barkan, "Divergence and Convergence in Kenya and Tanzania: Pressures for Reform," in
Joel D. Barkan, ed., Beyond Capitalism versus Socialism in Kenya and Tanzania (Boulder, Colo.: Lynne
Rienner, 1994).
somewhat more arid and less densely populated than the Kenyan dis-
trict, the two are similar along many other important characteristics.
Busia, Kenya, and Meatu, Tanzania, are located near each other
(roughly five hundred kilometers apart) on opposite sides of Lake Vic-
toria,19 both are overwhelmingly rural and have the same staple crops
(maize, sorghum, and cassava), although most of Meatu, Tanzania, has
only one harvest per year while Busia, Kenya, has two.20 The areas were
also part of a shared precolonial historical universe, with extensive mi-
gration across what is today the Kenya-Tanzania border.21 The current
border was drawn in the period 1886-90 by British and German colo-
nial authorities largely ignorant of the ethnic and political entities that
existed in the region,22 and it consists of a straight line drawn from the
western edge of Mount Kilimanjaro to the point where the shore of
Lake Victoria intersects one degree south latitude (the latitude at which
Britain and Germany had decided to split the lake roughly in half).
The arbitrary nature of African boundary creation during the colonial
period is at the heart of the empirical strategy of this article.
The total 1989 population of Busia was 401,658 and the current
population of Meatu is approximately 201,981. The two districts have
similar ethnic compositions, with majority Niger-Kordofanian (Bantu)
populations and substantial Nilo-Saharan minorities: the dominant
Luhya ethnic group (which is Bantu speaking) constitutes nearly 70
percent of the population in the Kenyan district, while the majority
Sukuma group (also Bantu) constitutes roughly 85 percent of the popu-
lation of the Tanzanian district.23 Armed conflict associated with cattle
raids was common in both areas during the precolonial period, and
interethnic relations were qualitatively similar.24 Another similarity lies
in the realm of language use: unlike many other regions of Tanzania,
Meatu district had minimal trade ties with Swahili speakers from the
" The ideal research design should not choose districts literally straddling a common border, since
border areas might be influenced by the neighboring country, and the existence of these "spillovers"
would complicate the interpretation of differences across the districts.
20
Average annual rainfall in Meatu and Busia is approximately 700 millimeters and 1500 millime-
ters, respectively.
21
David Lee Schoenbraun, A Green Place, a Good Place: Agrarian Change, Gender, and Social Identity
in the Great Lakes Region to the Fifteenth Century (Portsmouth, N.H.: Heinemann Press, 1998).
22
A. C. McEwan, International Boundaries ofEast Africa (Oxford: At the Clarendon Press, 1971).
11
Government of Kenya, Kenya Population Census, 1989 (Nairobi: Central Bureau of Statistics,
1994); Government of Tanzania, Shinyanga Region Socio-economic Profile (Dar es Salaam, 1999).
4
- See Daudi Kidamala and E. R. Danielson, "A Brief History of the Waniramba People up to the
Time of the German Occupation," Tanganyika Notes and Records 56 (1961); Paul Puritt, "Tribal Rela-
tions," in Fatoumata Agnes Diarra and Yash P. Ghai, eds., Two Studies on Ethnic Group Relations in
Africa: Senegal and the United Republic of Tanzania (Paris: UNESCO, 1974); and Government of Kenya,
Kenya Socio-Cultural Profiles: Busia District, ed. Gideon Were (Nairobi: Ministry of Planning and Na-
tional Development, 1986).
334 WORLD POLITICS
East African coast during the nineteenth century, so few residents of
the area spoke Swahili upon independence,25 and Swahili similarly was
not widely spoken in western Kenya during the precolonial period.26
None of the main ethnic groups in either district have been dominant
in national politics (for their respective countries) during the post-
independence period.
Moreover, community members are leading players in funding local
public goods in both countries, through school committees and water
committees in Kenya and through village councils in Tanzania, so it is
possible to compare local fund-raising across the two districts. Public
finance expenditures were increasingly decentralized in Tanzania dur-
ing the 1990s, although the central government continues to play the
leading role in paying teachers' salaries, as in Kenya.27
The East African Citizenship Project—which surveyed representa-
tive national samples of eight thousand schoolchildren in Kenya and
Tanzania in 1966-67—provides further evidence of commonalities.
Results from nearly identical survey instruments administered in both
countries on politics, citizenship, and ethnicity support the claim that
political attitudes were extremely similar. In fact, Koff and Von der
Muhll28 conclude that "there is an often startling similarity between the
responses given by Kenyan and Tanzanian students. . . . The cross-
national similarities are so constant as to raise questions about the sig-
nificance of the nation-state as a differentiating variable."29
Of course, this is not to say that Kenya and Tanzania were identical
at independence. Nairobi was the cosmopolitan capital of British East
Africa, with a growing industrial base, and Kenya had experienced a
much more violent path toward independence than Tanzania, most
dramatically illustrated in the so-called Mau-Mau uprising of the
1950s. Tanzania is also somewhat more ethnically diverse than Kenya
on the whole, although the differences are relatively minor.30 Nonethe-
less, many social scientists have taken the fundamental similarity of
25
M. H. Abdulaziz, "Tanzania's National Language Policy and the Rise of Swahili Political Cul-
ture," in W. Whiteley, ed., Language Use and Social Change (London: Oxford University Press, 1971),
171-72.
26
T P. Gorman, "The Development of Language Policy in Kenya with Particular Reference to the
Educational System," in W. H. Whiteley, ed., Language in Kenya (Nairobi: Oxford, 1974).
27
Ole Therkildsen, "Contextual Issues in Decentralization of Primary Education in Tanzania," In-
ternationalJournal ofEducational Development 20 (September 2000).
28
David Koff and George Von der Muhll, "Political Socialization in Kenya and Tanzania: A Com-
parative Analysis," Journal of Modern African Studies 5, no. 1 (1967), 50.
29
A limitation is that the dataset does not contain preindependence information, and since nation-
building policies had begun to diverge by 1967 (the year of the Arusha Declaration), it cannot serve as
a baseline.
30
For a discussion of preindependence differences, see Court and Kinyanjui (fn. 17).
and Tanzania as an analytical starting point, and this article fol-
lows in that tradition.31 Beyond East Africa, other scholars have also
used the colonial-era placement of other African national borders as
natural experiments to evaluate the impact of policies and institutions.32
POSTINDEPENDENCE DIVERGENCES
Despite these geographical, historical, and institutional commonalities
in western Kenya and western Tanzania, postcolonial central govern-
ments in the two countries have pursued radically different public poli-
cies toward ethnicity, and we argue that this divergence has had an
impact on current ethnic relations.
National language policy is an area of major differences. Barkan 33
writes that "the potential for [ethnic] conflict in Tanzania has . . . been
muted by the near universal use of Kiswahili, which replaced English
as the country's official language in the mid-1960s." Swahili (or
Kiswahili in East Africa) is an indigenous African language originating
on the Indian Ocean coast of East Africa and is seen as largely ethni-
cally neutral in both countries. The Tanzanian regime quickly pushed
for total Swahilization of government administration after independ-
ence and established the National Swahili Council to promote its use
in all spheres of public life.34 In contrast, although Swahili has long
been widely spoken in Kenya as a lingua franca, it competes with Eng-
lish and local "vernacular" languages such as Kikuyu in official settings,
including political forums and schools. For example, vernaculars—
rather than Swahili—typically serve as the medium of primary school
instruction through the fourth grade; after that English becomes the
principal language.
The public school curriculum in Tanzania has been aggressively employed
as a nation-building tool. The curriculum stresses common Tanzanian
history, culture, and values and inculcates students with a strong sense
of national and Pan-African identity.35 By the late 1960s political edu-
31
Barkan (fn. 18).
'2For instance, William F. S. Miles, HausalandDivided: Colonialism and Independence in Nigeria and
Niger (Ithaca, N.Y.: Cornell University Press, 1994); and Daniel Posner, "The Political Salience of
Cultural Difference: Why Chewas and Tumbukas Are Allies in Zambia and Adversaries in Malawi"
(Manuscript, Department of Political Science, UCLA, 2002).
"Barkan (fn. 18).
34
Edgar C. Polome, "Tanzania: A Socio-Linguistic Perspective," in Edgar C. Polome and C. P. Hill,
eds., Language in Tanzania (Oxford: Oxford University Press, 1980). David Laitin discusses language
policy and nation building in Africa, and the existence of lingua francas that, like Swahili in East
Africa, could be employed to strengthen national identities; see Laitin, Language Repertoires and State
Construction in Africa (Cambridge: Cambridge University Press, 1992)
33
David Court, "The Education System as a Response to Inequality," in Joel D. Barkan, ed., Politics
and Public Policy in Kenya and Tanzania (New York: Praeger, 1984). Mark Gradstein and Moshe
336 WORLD POLITICS
cation was included as a standard curriculum subject in both primary

and secondary school, and it was tested on national exams.36 Moreover,
by the 1970s all future teachers were required to serve in the paramili-
tary national service organization, which indoctrinated them in the
ideals of the regime.37 Prewitt, Von der Muhll, and Court asserted that
"Tanzania is unique among African nations in the extent to which it
has self-consciously sought to adapt the educational system inherited
at independence to the goals of the postcolonial leadership."38
Although the Kenyan Ministry of Education made several nation-
building pronouncements in the 1960s,39 Court and Ghai observed
that these were merely "vague invocations,"40 as "there [was] little evi-
dence within schools that the rhetoric [was] followed by any serious
attempts to make real changes."41 Nearly twenty years after independ-
ence, Court and Kinyanjui concluded that "Tanzanian students have a
stronger sense of national identity than their Kenyan counterparts."42
Unlike in Tanzania, the central government in Kenya has not used the
school curriculum to promote a coherent national linguistic or ideolog-
ical identity: the official geography, history, and civics (GHC) curricu-
lum does not study Kenya as a nation until grade 5. The focus on
provincial geography and history in grades 1-4 probably serves to exac-
erbate regional and ethnic divisions, especially among the many
Kenyans who drop out of primary school before grade 5.43
Another important component of the reform package carried out in
Tanzania was the complete overhaul of local government institutions, with
the aim of strengthening village councils and district councils. In
Kenya, by contrast, the colonial-era system of centrally appointed tribal
Justman present a formal political economy model in which political socialization through public ed-
ucation may promote social cohesion and economic growth: see Gradstein and Justman, "Education,
Social Cohesion, and Economic Growth," American Economic Review 92, no. 4 (2002).
36
Court and Kinyanjui (fh. 17), 67.
37
John White, "The Historical Background to National Education in Tanzania," in Edgar C.
Polome and C. P. Hill, eds., Language in Tanzania (Oxford: Oxford University Press, 1980).
38
Prewitt,Von der Muhll, and Court (fn. 16), 222.
39
Koff and Von der Muhll (fn. 28).
40
David Court and Dharam Ghai, eds., Education, Society, and Development: New Perspectivesfrom
Kenya (Nairobi: University of Nairobi Institute of Development Studies and Oxford University Press,
1974), 7.
41
Ibid., 19.
42
Court and Kinyanjui (fn. 17), 69.
43
Quantitative evidence from current schoolbooks also suggests that the Kenyan curriculum is con-
siderably less Pan-Afncanist in orientation than the Tanzanian curriculum. As a rough measure of cur-
ricular emphasis, we counted the number of times the word "Africa" (or "African") appears in nine
current Kenyan and Tanzanian GHC textbooks for grades 3—6 and found more than twice as many in-
stances in the Tanzanian textbooks (66.5 times per book) as in the Kenyan books (28.6).
chiefs has been retained. Kenya has no local government institution
comparable in authority to the elected Tanzanian village council. Tra-
ditional rural authorities and customary tribal law inherited from the
colonial period were completely dismantled in Tanzania upon inde-
pendence, and this may have played a role in further diminishing the
place of ethnicity in Tanzanian public life relative to that in Kenya.
Part of these policy divergences can be attributed to the personalities
andphilosophies of the respective independence leaders, Jomo Kenyatta and
Julius Nyerere. Inspired by a Pan-Africanist and socialist political phi-
losophy, the gifted Tanzanian leader Nyerere forcefully downplayed the
role of ethnic affiliation in public life and instead emphasized a single
Tanzanian national identity. A founding principle of Nyerere's ruling
TANU political party was "to fight tribalism and any other factors which
would hinder the development of unity among Africans."44
The nation-building role of the Kenyan central government could
not be more different. The first two postindependence presidents,
Kenyatta and Daniel arap Moi, are perceived within Kenya as "tribal-
ists," political opportunists who thrived on the politics of ethnic divi-
sion. Ethnicity has become the primary cleavage of political life in
Kenya—as in many other African countries—and the Moi regime was
widely implicated in arming and financing violent ethnic militias be-
fore national elections in 1992 and 1997. The clashes that were fo-
mented left hundreds dead.45
Finally, the regional distribution of central government resources—
for education, health, and infrastructure—has politicized ethnicity to a
far greater extent in Kenya. While the equitable regional distribution of
public investment in education, health, and infrastructure has been a
centerpiece of Tanzanian socialist policies since the 1960s,46 the post-
independence regime in Kenya heavily favored the ethnically Kikuyu
areas that formed the core of Kenyatta's political support, and the same
has been true of political appointments. After 1978 favoritism toward
ruling party areas continued, but shifted to reflect Moi's new ruling
coalition centered in the Rift Valley, as documented by Barkan and
Chege.47 This further contributed to the political salience of regional
44
M. H. Abdulaziz, "The Ecology of Tanzanian National Language Policy," in Edgar C. Polome
and C. P. Hill, eds., Language in Tanzania (Oxford: Oxford University Press, 1980).
43
Stephen Ndegwa, "Citizenship and Ethnicity: An Examination of Two Transition Moments in
Kenyan Politics," American Political Science Review 91, no. 3 (1997).
•"Court and Kinyanjui (fn. 17).
4
'Joel D. Barkan and Michael Chege, "Decentralising the State: District Focus and the Politics of
Reallocation in Kenya," Journal of'Modern African Studies 27, no. 3 (1989).
338 WORLD POLITICS
and ethnic identities. A recent article asserts that in Kenya "ruling
tribes get more than those out of power.... As [Moi] leaves office, the
nation is dead, only the tribe remains."48
Taken together, the various components of the Tanzanian reforms—
the promotion of Swahili as a national language, political and civic ed-
ucation in schools, the dismantling of tribal authorities, and the
relatively equal regional distribution of resources—contributed to the
growing strength of a coherent and popular national identity that binds
Tanzanians together across ethnic lines. The recent Afro-Barometer
public opinion surveys conducted among representative adult samples
in twelve countries during 1999-2001 (although not in Kenya, unfor-
tunately) provide further evidence that popular notions of ethnic and
national identity in Tanzania are in fact radically different from those
in other African countries. 49 When asked the open-ended question,
"Which specific group do you feel you belong to first and foremost,"
only 3 percent of Tanzanians responded in terms of an ethnic, language,
or tribal affiliation, the lowest of the twelve countries in the sample
with the exception of small and homogeneous Lesotho (at 2 percent).
Instead, 76 percent of Tanzanians answered in terms of an occupational
category (for example, farmer). This low rate of attachment to ethnic
identity stands in sharp contrast to other countries—Nigeria (48 per-
cent), Namibia (46 percent), Mali (39 percent), Malawi (38 percent),
and Zimbabwe (36 percent)—where, as in Kenya, ethnic divisions have
been politicized.50 Of the twelve countries surveyed, Tanzanians also
show among the highest levels of support for democracy, confidence in
government institutions, and trust in fellow citizens.51 Chaligha et al.
conclude:
President Nyerere's efforts to mould a national identity (for example, by em-
phasizing Kiswahili and abolishing traditional rule) have borne fruit.... If Tan-
zania was once an artificial construct of colonial mapmakers, it is no more. . . .
[T]he extent of common perception of nationhood and the lack of ethnic con-
siderations in politics is an important reason that Tanzania has been one of the
most politically stable countries in the region.52
These findings link up with the two theories of diversity and collec-
tive action presented in Section II. First, as the national Tanzanian
48
Dennis Onyango, "How Shall We Detribalise Presidency," Daily Nation (Nairobi), October 21,
2002.
"'Afro-Barometer Network, Afro-Barometer Round I: Compendium of Comparative Data from a
Twelve-Nation Survey, Afro-Barometer Paper, no. 11 (2002).
50
The other countries are Botswana, Ghana, South Africa, Uganda, and Zambia.
51
A. Chaligha, R. Mattes, M. Bratton, and Y. D. Davids, Uncritical Citizens or Patient Trustees? Tan-
zanians' Views ofPolitical and Economic Reform, Afro-Barometer Paper, no. 18 (2002).
52
Ibid., 11.
identity gains political salience, "taste" theories become less important,
since individuals increasingly identify with all citizens as fellow Tanza-
nians rather than just with their own tribe. They are thus willing to
fund public goods that benefit "other" groups. Second, to the extent
that the reforms also increase interethnic social interactions—for in-
stance, at the large public meetings regularly held in Tanzanian vil-
lages—they also increase the likelihood of stronger "social sanctions"
across ethnic groups, thereby reducing free riding and improving local
collective action.
IV. DATA
The empirical analysis utilizes household-level, school, and local gov-

ernment survey data collected in Kenya and Tanzania between 1996
and 2002.
DATA FROM BUSIA, KENYA

Detailed data for 100 of the 337 rural primary schools in Busia and
Teso districts were collected from pupil, school, and teacher question-
naires in early 1996 as baseline information for a nongovernmental or-
ganization (NGO) assistance project, and not for this study explicitly.53
A variety of financial and demographic data for these schools were col-
lected in 1996, and a pupil questionnaire was administered by trained
survey enumerators in grades 6-8, focusing on pupil schooling back-
ground, family educational characteristics and asset ownership, and self-
described ethnic affiliation. In total, 100 school questionnaires and 861
teacher questionnaires were also administered.54 NGO enumerators used
portable Global Positioning System (GPS) machines to collect latitude
and longitude for all primary schools and water wells in the sample.
The analysis below includes the 84 of 100 schools with complete data.
Ethnolinguistic fractionalization (ELF) is used as the principal em-
pirical measure of ethnic diversity in this study, mainly because this
measure has the advantage of comparability with the existing literature.
Ethnolinguistic fractionalization is the probability two people ran-
domly drawn from the population are from distinct groups, and is
closely related to a Herfindahl index; formally, ELF =1-2. (Propor-
53
In 1996 the original Busia district was split in two: Teso district is the northern part of the origi-
nal district, and Busia district the southern part; for simplicity, the combined area is referred to as
"Busia."
''School questionnaires—filled out by schoolmasters with the assistance of a trained enumerator—
contain detailed information on schoolfinances,infrastructure, inputs, and pupil enrollment.
340 WORLD POLITICS
2
tion of group i) . In Kenya the principal measure of "local" ethnic di-
versity for a primary school is computed among all pupils located
within five kilometers of that school, where all distances are determined
using the GPS data. The five-kilometer radius appears to be a rough
upper limit on the distance that children may walk to attend school on
a daily basis and thus on what may constitute a school "community," al-
though the empirical results are robust to radiuses of three to six kilo-
meters (results not shown). These data were created from 1996
government District Education Office examination name lists.
Both the central government and local school committees play im-
portant roles in Kenyan primary school finance: while the national
Kenya Ministry of Education pays teacher salaries, school committees
raise funds locally for desks, classrooms, books, and chalk. Although
the teacher salaries and benefits paid by the central government account
for most spending—approximately 90 percent—a reduction in local
funding could have an important impact to the extent local inputs and
teachers are complements in educational production. Each primary
school is managed by its own committee, which is composed of class
representatives directly elected by parents.
Parents raise local school funds through two mechanisms: school
fees and local fund-raising events. Annual school fees are set by the
school committee and collected by the headmaster. The other impor-
tant source of local primary school funding in western Kenya, account-
ing for approximately one-third of local funding, are village
fund-raisers called harambees. At these events parents and other com-
munity members publicly pledge financial support for a school project
(for example, classroom construction). While contributions at these
events are supposedly voluntary, school committees often announce ex-
pected contributions for parents, and actual contributions may be
recorded by the committee. The main school finance outcome for
Kenya in this study is total local school funding collected per pupil in
1995 from both fees and harambees, while school facilities and inputs—
the number of desks per pupil, latrines per pupil, and classrooms per
pupil in 1996—are other outcome measures.
Water wells are another important local public good in rural East
Africa, since well water is generally safer to drink than alternative water
sources, such as stream or lake water, and the lack of safe drinking
water is a major public health problem that contributes to the spread of
such diseases as amebiasis, cholera, and schistosomiasis.55 The vast ma-
55
Government of Kenya (fn. 24).
jority of community wells in western Kenya were constructed in
1982-91 with the assistance of the Finnish government, through an or-
ganization called the Kenya-Finland Development Cooperation
(KEFINCO). KEFINCO identified well sites in cooperation with commu-
nities, dug the boreholes, and provided the equipment for operating the
wells. Communities were then responsible for forming well committees
in charge of maintenance and collecting usage and repair fees from
local residents. The committees operate on a voluntary basis with little
explicit public authority for revenue collection, so, as with school funds,
their ability to collect fees largely depends on their success in exerting
social pressure in the local community.
The data on well maintenance were collected for this study in a sur-
vey of 667 wells conducted in Kenya from October 2000 to August
2001 by NGO field-workers for this study. The sample consists of the
universe of modern borehole wells constructed between 1982 and 1991
by KEFINCO. The current condition of the wells thus reflects the success
of local collective action in maintenance from the 1980s through 2001.
The principal dependent variable for well maintenance is an indicator
variable that takes on a value of one if water flow was judged "normal"
by field-workers, and zero if either no water flows or if water flow is
"very low." Only 56 percent of the wells had "normal" water flow at the
time of the survey, suggesting widespread collective action failures.
In the empirical analysis for Kenya, the unit of observation is a "pri-
mary school community"; we consider all wells within five kilometers
of the school as "assigned to" that primary school. Unlike Tanzania,
rural Kenya does not have coherent villages with fixed boundaries, and
for this reason the Kenya analysis focuses on school communities; this
is also necessary in order to perform joint tests across the school and
well outcomes, as described below.
DATA FROM MEATU, TANZANIA

Data collection was carried out in all seventy-one villages in Meatu,
Tanzania, during 2001-2 by NGO field staff. The surveys were designed
to be comparable with the existing survey data from Kenya, but consid-
erable additional information was also collected. The village council
survey data relied both on interviews with village council members and
on local administrative records—especially the tax register. Specifically,
we collected retrospective information on all village public goods proj-
ects by year from 1997 to 2002, including contributions from commu-
nity members as well as outside funding from government agencies and
NGOs. Field-workers also observed the current condition of primary
342 WORLD POLITICS
schools and water well maintenance using the same survey instruments
employed to collect the Kenya data, as well as information on road and
health clinic infrastructure. They recorded total village population, and
they determined ethnic composition by gathering the ethnic affiliation
of individuals in a 20 percent random sample of the tax register (with
the assistance of local officials, enumerators counted off each fifth name
in the tax register for inclusion in the sample). The tax register includes
all village adult males; unfortunately there are no comparable data for
females.
Once the village council has decided on a local project and set con-
tribution levels, local contributions are collected in cooperation with
subvillage chairs (each village contains six to ten subvillages) and the
leaders of the local elders council. Each household is obliged to make a
standard contribution, which may be in cash or in kind, usually as ma-
terials and labor. Some funding for local projects in Tanzania also
comes from the Meatu District Council (MDC), other Tanzanian gov-
ernment agencies (for example, the Tanzanian Social Action Fund,
TASAF), or NGOs in a few cases, and this assistance is typically structured
as cost sharing: for example, the MDC provides assistance for classroom
construction provided that the village council raises a certain portion of
the costs locally. NGOs often follow similar cost-sharing policies both in
Meatu, Tanzania, and in Busia, Kenya; hence, in both districts the bulk
of funding, materials, and labor for public goods is raised locally, with
some limited outside assistance. The local public finance outcomes de-
scribed in this article thus mainly capture the ability of communities to
raise funds and supplies locally, as well as their ability to secure some
additional funds from other donors. Both are important collective ac-
tion outcomes in their own right.
We also collected information on social capital, including local com-
munity groups, attitudes regarding trust and cooperation, and atten-
dance at village meetings (total attendance at all 2001 meetings per
household). Village meetings are held for local elections, to discuss de-
velopment projects, and to disseminate information from higher levels
of government (for example, to promote HIV/AIDS awareness), and they
serve as the focal point for local politics.
The 2001-2 household surveys were administered to approximately
15-20 households from each village, with 1,293 households surveyed
in all. Households to be surveyed were randomly chosen from the tax
register, as were neighbors of the register households. The household
survey included detailed socioeconomic, migration, and demographic
questions and a consumption expenditure module for a subset of
households, and these data are used to create village controls. Five vil-
lages are missing at least some survey data, reducing the final sample to
sixty-six villages.56
V. IDENTIFYING E T H N I C DIVERSITY EFFECTS
There are two steps in the econometric identification strategy. First, we

estimate the impact of ethnic diversity on local collective action out-
comes in both Kenya and Tanzania. Second, we argue that differences
across the districts—in terms of the impact of ethnic diversity on local
outcomes—were most likely caused by divergent nation-building poli-
cies, rather than by other factors.
ETHNIC DIVERSITY IN KENYA
Busia and Teso districts in Kenya are moderately ethnically diverse: the
largest ethnic groups are the Luhya (67 percent of the sample), Teso (26
percent), and Luo (5 percent). The Luo and Teso are Nilo-Saharan
ethnolinguistic groups with pastoralist traditions, and the Luhya are a
Bantu (Niger-Kordofanian) group. Luhyas are the majority ethnic
group in southern Busia district and Tesos are numerically dominant in
the north, although there are minority communities spread throughout
the area.
The main concern regarding econometric identification is the possi-
bility that it could be local unobservable characteristics correlated with
ethnic diversity in each district—rather than ethnic diversity itself—
that are in fact driving the estimated diversity effects. The exogeneity
of ethnic land-settlement patterns forms a basis for the empirical iden-
tification strategy in Busia, as a variety of evidence suggests that cur-
rent levels of local ethnic diversity in Busia district is largely the product
of historical accident rather than recent migration: "The nineteenth cen-
tury was a time of considerable unrest throughout the District, with
conflict between the Luhya groups, Luo, Teso and Kalenjin."57 Were
writes that "various factors—famine, epidemics, domestic disputes, the
spirit of adventure and warfare—made the inhabitants of the region ex-
tremely mobile" from the seventeenth to the nineteenth century.58
56
Ethnic diversity (ELF) in the villages with missing data is nearly identical on average to diversity
in villages that are included in the sample, and the small difference in ELF (0.16 to 0.13) is not statis-
tically significantly different than zero at traditional confidence levels. Hence, even if thefivevillages
with some missing data happened to be poor performers in terms of local collective action, the inclu-
sion of these villages would be unlikely to substantially change the main empirical results.
s7
38
Gideon Were, A History ofthe Abaluyia of Western Kenya (Nairobi: East African Publishing, 1967).
344 WORLD POLITICS
The emergence of British colonial authority in western Kenya in
1894 brought an end to wars and cattle raiding, as well as to the large
movements of populations that accompanied them. Morgan and Shaf-
fer write that ethnic land claims were "frozen by the Colonial Govern-
ment by the demarcation of'African Land Units.' This prevented the
expansion of tribes into another's territory and thus eliminated the
principal source of major inter-tribal wars."59 Land demarcation and in-
dividual land registration during the postcolonial period "has frozen the
previously fluid situation and virtually halted the traditional mobility."60
Busia was free of European settlement—and resulting disruptions of
land claims—during the colonial period.
Comparing residential ethnic composition at the geographic division
level in 1996 (using pupil survey data) with residential composition in
1962 (using Kenyan census data)61 suggests that ethnic residence pat-
terns have been largely stable: the ordering of residential ethnic diver-
sity across geographic divisions is identical in 1962 and 1996 (results
not shown). Recent survey evidence also suggests that land sales and
residential mobility are extremely rare.62 Although residential patterns
in this area are stable, households can choose which primary school
their children will attend and which water well to use, creating endoge-
nous populations of schoolchildren and water users. To limit bias due
to endogenous sorting within walking distance of the household, we
employ ethnic diversity within five kilometers of each school as our
principal local diversity measure in the Kenya analysis, rather than the
diversity of actual pupils or well users.
ETHNIC DIVERSITY IN TANZANIA
An empirical methodology similar to that used in the Kenya analysis is
used to estimate the relationship between local ethnic diversity and
provision of public goods in Meatu, Tanzania. As in Kenya, under-
standing patterns of land settlement is central to the econometric iden-
tification strategy. Meatu district was relatively sparsely populated until
the mid-twentieth century, after which increasing numbers of ethni-
cally Sukuma, Taturu, and Nyiramba individuals from neighboring
areas migrated there in search of farmland, so unfortunately, unlike
59
W. T. W Morgan, and N. M. Shaffer, Population of Kenya, Density and Distribution: A Geographi-
cal Introduction to the Kenya Population Census, 1962 (Nairobi: Oxford University Press, 1966).
60
"Government of Kenya, Kenya Population Census, 1962 (Nairobi: Central Bureau of Statistics,
1965).
"Miguel and Gugerty (fn. 6).
Busia, Kenya, there is no compelling historical migration "natural ex-
periment" that can be used to identify effects of ethnic diversity. Short-
run endogenous local sorting is less salient in Meatu, however, since
individuals live in separated villages and population density is lower;
hence, households generally have fewer schools and wells to choose from.
The relative stability of residential patterns helps rule out the most
obvious forms of recent sorting in response to either socioeconomic or
public good variation. There was local migration associated with the
forced villagization program of the mid-1970s (described below), but
there is no evidence villagization per se significantly altered local ethnic
diversity since individuals seldom moved more than a few miles from
their original homes.63 Stringent residency regulations during the so-
cialist period further dampened migration, and, as in rural Kenya, the
absence of a well-functioning land market currently contributes to low
rates of mobility. Survey evidence also indicates that residential patterns
have in fact been largely stable in Meatu recently: to illustrate, over 80
percent of the young adult (under thirty years) male respondents in the
sample have been living in the same village for at least ten years, and as
a further check, rates of residential stability are also nearly identical in
relatively high ethnic diversity {ELF > 0.15) and low diversity (ELF<,
0.15) villages, at 80 and 83 percent, respectively.64
Moreover, to the extent that there is endogenous sorting in Meatu,
Tanzania, the sorting bias is likely to be negative. There is a widespread
view that diverse areas—mainly in Nyalanja division—are marginalized
and have relatively poor-quality land, and the unconditional correlation
between village per capita income and diversity is in fact negative (al-
though not statistically significant; results not shown). To the extent
that land quality is unobserved and is not entirely captured by other
controls, this would negatively bias estimated diversity effects, insofar
as poorer villages are less successful in the provision of public goods.
The above arguments do not definitively eliminate omitted variable
bias as a potential problem, and it remains the central concern with the
identification strategy in Meatu district. Yet the fact that omitted vari-
able bias is likely to be negative argues against the existence of negative
diversity effects in this case, since estimates of ethnic diversity effects
in Meatu are near zero or even positive, as discussed below.
"Juma Volter Mwapachu, "Operation Planned Villages in Rural Tanzania: A Revolutionary

Strategy for Development," Mbioni 7, no. 11 (1975).
64
It is natural to focus on male residential stability. Since marital exogamy is practiced in this re-
gion, most women move to their husband's home upon marriage.
346 WORLD POLITICS
IDENTIFYING T H E IMPACT O F N A T I O N BUILDING

T h e two main methodological weaknesses of the cross-district com-
parison are, first, the small sample size of only two countries and, sec-
ond, the lack of longitudinal data on collective action outcomes, which
would greatly strengthen the case that the two districts were in fact
largely comparable in the 1960s and have since diverged. However,
these methodological weaknesses are impossible to overcome at this
time, given the lack of internationally comparable data on ethnic poli-
cies, historical patterns of ethnic relations, and current local public
goods outcomes.
Unfortunately, there is no quantitative evidence (that we are aware
of) on interethnic cooperation in these districts in the preindependence
period. But as there is also no compelling evidence suggesting that eth-
nic relations were dramatically different in the two areas, we maintain
the assumption that current interethnic cooperation in the two districts
would have been largely similar in the absence of the nation-building policy
divergences described in Section III.
A third concern is that preexisting ethnic relations endogenously af-
fected the nation-building policies that were ultimately chosen, such
that causality actually runs from ethnic cooperation to nation-building,
rather than vice versa. Although the nation-building policies chosen in
Kenya and Tanzania, as well as the characteristics of postindependence
leaders, may indeed have been related to the nature of average ethnic
relations at the national level in both countries, all that is necessaryfor a
valid comparison of the impact of nation-building policies on Busia, Kenya,
and Meatu, Tanzania, is that the choice of national policies was not directly
related to ethnic relations in these two small and politically marginal dis-
tricts, and this seems plausible.
The study is also unable to separately estimate the effects of various
components of the Tanzanian nation-building package, in language,
education, and institutional reform. These components may in prin-
ciple interact in complex and multiple ways, and we are entirely unable
to estimate these interactions in this study. Instead, the effects pre-
sented below should be seen as the estimated impact of the entire Tan-
zanian reform package on local collective action in diverse areas,
relative to Kenya.
A reading of the recent history of western Kenya and western Tan-
zania indicates that differences in current levels of interethnic coopera-
tion across Busia, Kenya, and Meatu, Tanzania, are most likely due to
their strikingly different nation-building policies during the post-
colonial period, rather than to other factors. Other than the nation-
building policies described above, the most sustained policy divergences
between Busia, Kenya, and Meatu, Tanzania, occurred during the
1970s: from August 1974 through 1977 the Shinyanga regional gov-
ernment pursued a policy of forced villagization in which over 340,000
rural residents were compelled to leave their homes and move to nearby
villages, sometimes by force—including the burning of resistors'
homes.65 The centerpiece of Tanzanian socialism was the goal of con-
centrating Tanzania's scattered populations into Ujamaa villages, where
government could, in theory, more efficiently provide public services
and where collectivized farming would take place.66 Nationally, during
this period of radical reform, Tanzanian economic growth rates lagged
far behind Kenyan rates.67
If anything, however, the policies of this period appear likely to have
inflamed ethnic tensions rather than promoting cooperation: the non-
Bantu Taturu and Hadzabe minority ethnic groups were particularly
hard hit by villagization, since it contributed to the erosion of their tra-
ditional seminomadic lifestyles. To the extent therefore that ethnic re-
lations are currently better in western Tanzania than in western Kenya,
it is unlikely to be due to the arbitrary and violent villagization poli-
cies—and anemic economic growth—of the Ujamaa period. Yet it is
impossible to rule out definitively that villagization per se improved in-
terethnic cooperation in Tanzania over the long run by promoting fre-
quent social interactions among members of different ethnic groups in
the new villages.68
Two other considerations that might account for differences between
western Kenya and western Tanzania merit discussion. First, if greater
prosperity allows communities to partially overcome ethnic divisions in
local fund-raising, this would imply that diverse communities in the
Kenyan district would likely have somewhat better outcomes than in
the Tanzanian district—but we find the opposite, as discussed below,
65
Mwapachu(fn. 63).
"Dean E. McHenry, Jr., Tanzania's Ujamaa Villages (Berkeley: Institute of International Studies,
University of California, Berkeley, 1979).
67
Barkan (fn. 18).
68
An alternative explanation for the existence of a stronger national identity in Tanzania could be
the successful 1979 war repelling a Ugandan invasion. Although victorious wars have long been cred-
ited with promoting national unity, this hypothesis appears unlikely here for at least two reasons: first,
the Uganda war lasted only three months, and second, the war led to an exhausting six-year occupation
of Uganda that nearly bankrupted the country and contributed to the financial crisis of 1982; see D. F.
Gordon, "International Economic Relations, Regional Cooperation, and Foreign Policy," in Joel D.
Barkan, ed., Beyond Capitalism versus Socialism in Kenya and Tanzania (Boulder, Colo.: Lynne Rien-
ner, 1994).
348 WORLD POLITICS
suggesting that our estimated effects may be lower bounds. A second

possible channel is different levels of foreign aid to Kenya and Tanza-
nia. However, the focus of this study is local community fund-raising
and there is little external assistance to these districts.69
ECONOMETRIC ESTIMATION EQUATION

The basic empirical specification is presented in Equation 1. Y* is the
outcome measure, where k may denote local school funding, school in-
frastructure quality, well maintenance, or another outcome. ELFu is the
measure of local ethnic diversity (ethnolinguistic fractionalization),
where i denotes a community (within five kilometers around a school for
Kenya, and a village for Tanzania), and c denotes the country. Xk is a
vector of local socioeconomic, demographic, and geographic controls:70
Y? = a* + X.' pk + TkELF. + uk (1)

IC IC ^ IC IC ^ '
Regression disturbance terms are assumed to be independent across

geographic zones (Kenya) or wards (Tanzania) but are clustered within
the zones or wards (results are similar when disturbance terms are al-
lowed to be spatially correlated across schools using the method in
Conley,71 results not shown).
Equation 2 presents the empirical estimation framework for the
cross-district comparison, where data are pooled from both countries.
An indicator variable is included for Kenyan communities {KENYA..)
to capture any differences in average outcomes across the two districts
(for instance, due to differences in local fund-raising institutions or in
average socioeconomic status across Busia and Meatu), and this term is
also interacted with the vector of local controls:
Yh = a\ + ai KENYA. + X. 'j8f + {X * KENYA.)'B*L

u 1 2 ic tc r" 1 <- ,c icJ"2
k
+ x\ELF. + x\ {ELF * KENYA.} + u (2)
1 ic 2 *- tc ICJ r"u v
'
captures the impact of ethnic diversity on local outcomes in Tanzania

for outcome k, while x\+ xk2 is the effect in Kenya. The main hypothe-
69
Only six of eighty-four sample schools report having received over U.S.$100 in outside funding in
the 1996 survey, and U.S.S100 is a low level for schools with three hundred to four hundred students
on average. Similarly, a minority of projects in the Tanzanian district received any external central gov-
ernment or NGO assistance, and in many cases the amounts received were modest.
70
Unfortunately, there are no data on income (or consumption expenditures) for the Kenyan vil-
lages, so income cannot be included as a control variable in the main regressions.
''Timothy Conley, "GMM Estimation with Cross Sectional Dependence," Journal of Econometrics
92, no. 1 (1999).
k
sis of this article can be restated as HQ: T 2 = 0, jointly for all outcomes
k. Rejecting this hypothesis means that ethnic diversity has a signifi-
cantly different effect on local collective action in the two districts
under study, and this can be interpreted as the difference between the
effects of the exceptionally serious Tanzanian nation-building policies
versus policies in Kenya that may have exacerbated ethnic divisions.
The existence of multiple collective action outcomes provides addi-
tional statistical power to reject the hypothesis that the impact of di-
versity is the same in the two districts, since outcomes for a given
village are only imperfectly correlated due to various idiosyncratic fac-
tors, for instance, the competence of local committee officials who deal
with schools and water. To illustrate, the correlation between desks and
classrooms per pupil in Busia, Kenya, is approximately 0.4, while the
correlation between school funding per pupil and the quality of well
maintenance is positive but low, at only 0.1; similar patterns hold in
Meatu, Tanzania. So the confidence interval around the estimated im-
pact of diversity when data are pooled across outcomes is considerably
narrower than the interval for any single outcome. We use seemingly
unrelated regression (SUR), in which disturbance terms are allowed to
be correlated across outcome measures for a village (or school) during
hypothesis testing, to test whether the overall effect of ethnic diversity
differs across Busia and Meatu.
SUR regression coefficient estimates are identical to OLS here since
the explanatory variables are the same for all outcomes.72 The advan-
tage of SUR lies in allowing us to perform joint hypothesis tests utilizing
information across all five regressions. When there is limited correla-
tion in village disturbance terms across the different dependent vari-
ables, this method is equivalent to an increase in sample size—and this
explains why statistical significance may be higher for SUR hypothesis
tests than for any single OLS coefficient on its own.
VI. EMPIRICAL RESULTS FROM WESTERN KENYA AND

WESTERN TANZANIA
DESCRIPTIVE STATISTICS
Average levels of local ethnic diversity are similar in the two districts,
although somewhat higher in the Kenyan district: the average level of
ELF in Busia, Kenya, is 0.23 and in Meatu, Tanzania, 0.13 (see Table 1,
panel A). Nonetheless the supports of the two local ELF distributions
-2
' Jeffrey M. Wooldridge, Econometric Analysis of Cross-Section and Panel Data (Cambridge: MIT
Press, 2002).
TABLE 1
DESCRIPTIVE STATISTICS
1Weatu District, Tanzania Busia District, Kenya
Mean Std 06s. Mean Std 06s.
Panel A: Data for Tanzania and Kenya Dev. Villages Dev. Schools
Local ethnic fractionalization (ELF) 0.13 0.16 66 0.23 0.14 84
Average years of education 4.1 1.1 66 7.4 1.3 84
Proportion formal sector employment 0.06 0.07 66 0.23 0.10 84
Proportion of homes with iron roofs 0.26 0.20 66 0.24 0.10 84
Proportion households grow cash crops 0.61 0.25 66 0.40 0.24 84
Proportion households own cattle 0.47 0.17 66 0.60 0.19 84
Proportion Catholic 0.17 0.12 66 0.58 0.22 84
Annual expenditures on local primary 4.88 3.90 66 3.45 2.23 84
school projects per pupil (U.S.$)
Desks per primary school pupil 0.19 0.09 66 0.21 0.12 84
Latrines per primary school pupil 0.011 0.005 66 0.016 0.013 84
Classrooms per primary school pupil 0.013 0.005 66 0.031 0.014 84
Teachers per primary school pupil 0.013 0.004 66 0.039 0.015 84
Proportion wells with "normal 0.57 0.37 66 0.56 0.14 84
water flow"
Panel B: Datafor Tanzania

Number of households per village 413.4 178.8 66
Annual per capita consumption 198.4 81.2 66
expenditures (U.S.$)
Gini coefficient of annual per capita 0.36 0.15 66
consumption expenditures (at
village level)
Annual local expenditures on all 8.65 6.39 66
public goods projects, per household
(U.S.$)
Annual local expenditures on health 1.51 1.78 66
and water well projects, per household
(U.S.S)
Annual local tax collection, per 2.14 3.47 66
household (U.S.S)
Average number of completed local 0.67 0.40 66
public goods project, per year
Average household spending on local 12.3 14.5 66
taxes and school expenses (U.S.$)
[HH Survey]
Wells with normal water flow, per 0.008 0.009 66
household
Average road quality (scale 1-4) 2.5 0.8 65
Total community groups, per 0.026 0.017 66
household
TABLE 1 (cont.)
Meatu District, Tanzania Busia District,Kenya
Mean Std Obs. Mean Std Obs.
Panel B: Datafor Tanzania Dev. Villages Dev. Schools
Community group memberships 2.25 0.93 66
[HH survey]
Proportion survey respondents who 0.33 0.23 66
are community group members
[HH Survey]
Village meeting attendance, per 1.22 1.03 66
household
Proportion households attending a 0.94 0.07 66
village meeting [HH Survey]
In general, can you trust people in 0.83 0.11 66
other tribes?
Spirit of cooperation across tribes in 0.59 0.11 66
village, proportion stating "above
average"
Village unity, proportion stating 0.65 0.14 66
"above average"
Nation is more important than tribe 0.40 0.14 66
to respondent
SOURCES: Busia, Kenya, data are from the 1996ICS School and Pupil Questionnaires, 1996
Government Examination Namelists, and Global Positioning Systems (GPS) readings taken
by NGO field-workers. Each Kenyan data point refers to a primary school or to the five
kilometer radius around a school (in the case of the ethnic composition and water well
maintenance). Meatu, Tanzania, data are from the 2001-2 household village council sur-
veys, and ethnicity measures are for the village, computed using a 20 percent random sam-
ple of the village tax register.
For "Annual local expenditures on primary school projects per pupil (U.S.S)," the
Kenyan data are from school records about parent and community contributions to the
school in 1995. For Tanzania, local project values were obtained from the Meatu District
Council and the Dutch Rural Development Programme, and then combined with 2001-2
village council survey information on the types of project completed in each village to deter-
mine the overall value of local project activity. Thus measures for the two countries are not
entirely equivalent. All dollar figures for both countries are in 2002 U.S.S.
Local characteristics (education, employment, etc.) for Kenya are from the 1996 Pupil
Questionnaire, and are data about pupils' fathers. Local characteristics for Tanzania are from
the 2001-2 household survey, in which both men and women were surveyed, though two-
thirds of respondents were men. The gender of respondents may partially explain differences
in average reported socioeconomic characteristics between Busia, Kenya, and Meatu, Tanza-
nia, since educational attainment and formal sector employment are higher among men in
both countries. In addition, the Tanzanian sample contains both men and women, while
the Kenyan sample contains only men (fathers); some Tanzanian respondents were also eld-
erly, and the elderly tend to have less education and formal employment.
352 WORLD POLITICS
are nearly identical, ranging from 0 to 0.6, and there is considerable
variation in local ethnic diversity in both districts. Busia, Kenya, is sim-
ilar to Meatu, Tanzania, along certain socioeconomic characteristics—
including the proportion of homes with iron roofs, that grow a cash
crop, and livestock ownership—but is better off along several others.
For example, both average educational attainment and the proportion
of respondents with formal sector employment are substantially higher
in Busia, Kenya, than in Meatu, Tanzania.73 These socioeconomic char-
acteristics, as well as the proportion of Catholic households, are in-
cluded in all specifications as control variables.74
In terms of local public goods, school fund-raising levels are some-
what higher in Meatu, Tanzania, but the quality of school infrastruc-
ture in Busia, Kenya, is better along certain dimensions. For example,
while there are only 0.013 classrooms per pupil in Tanzania, there are
more than twice as many in Kenya and there are also differences in the
provision of latrines and in the pupil-teacher ratio across the two dis-
tricts, with Kenya having better infrastructure—most probably due to
its somewhat higher average socioeconomic level. Yet the proportion of
wells with "normal water flow" is nearly identical, at 57 percent in Meatu,
Tanzania, and 56 percent in Busia, Kenya, suggesting pervasive local col-
lective action failures with respect to water supplies in both districts.
Table 1, panel B, presents data that exist for the Tanzanian district
but not the Kenyan district, including average total annual expenditures
on local projects. Tanzanian villages funded U.S.$8.65 worth of local
public finance projects per household per year on average from 1997 to
2002, and there was considerable variation across villages (standard de-
viation 6.39). Most of the funding was spent on education, health, and
water projects, and villages completed 0.67 projects per year.
73
These differences are somewhat misleading: the Kenyan data are for pupils' fathers, while, even
though two-thirds of Tanzanian respondents were male and young adults—and thus comparable to
the Kenyan respondents—some Tanzanian respondents were women or elderly. Women and the elderly
have less education and formal employment than young men on average, and thus actual socioeconomic
gaps between the two districts are likely to be somewhat smaller than those reported in Table 1.
'" Religious diversity is not included as an explanatory variable in the analysis since local religious af-
filiation is not plausibly exogenous due to extensive missionary activity in both districts. Correlations
between religious fragmentation and local outcomes would be misleading if evangelical activity is most
successful in the poorest areas or in areas with low levels of social capital. The numerical strength of
"traditional" religions in Meatu—over 60 percent of the sample—also complicates the interpretation of
the religious fragmentation index, since it is difficult to disentangle different traditional belief systems
from ethnicity. Finally, since the most politically salient religious cleavage in East Africa is that be-
tween Christians and Muslims, the absence of large Muslim populations in these districts blunts the
most likely source of religious divisions.
ETHNIC DIVERSITY AND LOCAL COLLECTIVE ACTION OUTCOMES

IN KENYA AND TANZANIA
The two key terms in Table 2 are the coefficient estimate on the
ELF*KENYA interaction term, which is the difference between the im-
pact of ethnic diversity on the public goods outcome in Kenya versus
Tanzania, and the coefficient estimate on local ethnic diversity (ELF),
which can be interpreted as the relationship between ethnic diversity
and the local public goods outcome in Tanzania. The sum of these two
coefficient estimates is the impact of ethnic diversity in Kenya (the bot-
tom row of Table 2).
The estimated relationship between ethnic diversity and local provi-
sion of public goods in Busia, Kenya—the sum of the coefficient esti-
mates on the ELF and ELF*KENYA terms—is negative for all five
local public goods outcomes on which there are data for both countries
(local primary school funding per pupil, desks per pupil, latrines per
pupil, classrooms per pupil, and the proportion of water wells with nor-
mal flow), and statistically significantly different than zero for school
funds per pupil and desks per pupil. The school funds result implies
that the change from ethnic homogeneity to average levels of diversity
in Busia, Kenya, is associated with a drop of approximately 25 percent
in average funding—a large effect—while the estimated effect for
Meatu, Tanzania, is positive and statistically insignificant. Figure 2a
graphically presents the negative relationship between ethnic diversity
versus local school funding in Busia, Kenya, and Figure 2b presents the
relationship for desks per pupil.
In Miguel and Gugerty,75 each well is considered a separate data
point, and an ethnic diversity measure specific to that well (typically di-
versity within five kilometers of the well) is constructed. By way of con-
trast, in Table 2 we examine well maintenance within five kilometers of
each primary school and consider the local diversity measure of that pri-
mary school as the key explanatory variable. This school diversity measure
is a noisy measure of the ethnic diversity of each well, and the resulting
attenuation bias is the most likely explanation of why the results in Table
2 are weaker than in Miguel and Gugerty, where the comparable coeffi-
cient estimate on local diversity in Busia is -0.26 (standard error 0.14).
In Meatu, Tanzania, the coefficient estimate on ELF is positive for
all four primary school outcomes (and statistically significant for la-
trines per pupil), suggesting that there is not a negative relationship
75
Miguel and Gugerty (fn. 6).
TABLE 2
ETHNIC DIVERSITY AND LOCAL PUBLIC G O O D S : KENYA AND TANZANIA 3
Annual Proportion H0:f5 = 0

School Wells -with F-statistic
Spending/ Desks/ Latrines/ Classrooms/ Normal p-value
Pupil, U.SJ Pupil Pupil Pupil Flow (SUR)h
Explanatory Variable (1) (2) (3) (4) (5)
Local ethnic -7.7 -0.40** -0.014 -0.014 0.20 0.02**
diversity (ELF) * (5.9) (0.15) (0.013) (0.012) (0.31)
Kenya Indicator
Local ethnic 4.1 0.08 0.007** 0.006 -0.26 0.44
diversity (ELF) (5.6) (0.09) (0.003) (0.005) (0.24)
Kenya indicator -4.1 -0.08 0.025** 0.024** -0.43 0.03**
variable (3.6) (0.15) (0.012) (0.009) (0.26)
Socioeconomic Controls
Average years 0.52 0.013 0.0013**'" 0.0013* -0.083** 0.08*
of education (0.55) (0.011) (0.0004) (0.0007) (0.037)
Proportion formal -11.0 0.30* 0.015** 0.016 -0.31 0.38
sector employment (9.0) (0.17) (0.006) (0.010) (0.58)
Proportion homes -1.9 -0.05 -0.006*** -0.002 0.12 0.82
with iron roofs (2.3) (0.07) (0.002) (0.004) (0.18)
Proportion households -0.8 -0.03 0.000 -0.001 -0.12 0.96
grow cash crops (2.2) (0.04) (0.002) (0.002) (0.18)
Proportion households -2.6 -0.05 0.011*** -0.002 -0.27 0.37
own cattle (2.5) (0.05) (0.003) (0.005) (0.24)
Proportion Catholic 1.9 -0.06 -0.003 -0.011 -0.64 0.37
(3.3) (0.09) (0.003) (0.009) (0.52)
Socioeconomic Yes Yes Yes Yes Yes
controls * Kenya
Indicator
R2 0.15 0.19 0.13 0.41 0.19
Root MSE 3.07 0.098 0.011 0.011 0.25
Number of observations 150 150 150 150 150
Ethnic diversity -3.6* -0.32** -0.007 -0.008 -0.06 0.02**
effect, Kenya (2.0) (0.12) (0.012) (0.010) (0.19)
" Huber robust standard errors in parentheses. Significantly different than zero at 90% (*), 95% (**),
99% (***) confidence. Regression disturbance terms are clustered at the zone level for Kenya, and at
the ward level for Tanzania. The data contain eighty-four primary schools in Busia, Kenya, and sixty-
six villages in Meatu, Tanzania.
b
The hypothesis that the coefficient estimate on each term is equal to zero across the five outcomes
in the table is tested using SUR in the final column.
' School Funding Kenya (USD) Fitted values * Desks per pupil Kenya Fitted values
'.••••.•
.2 .4 .2 .4 .6
(a) Local Ethnic Diversity (ELF) (b) Local Ethnic Diversity (ELF)
Busia, Kenya: total local primary school Busia, Kenya: desks per primary school
funds per pupil (2001 U.S. dollars) in 1995 pupil in 1996 versus local ethnolinguistic
versus local ethnolinguistic fractionalization fractionalization
• School Funding Tanzania (USD) • Fitted values * Desks per pupil Tanzania Fitted values
.6-1
20-
.2-
0-
0 .2 .4 .6 0 .2 .4 .6
(c) Local Ethnic Diversity (ELF) (d) Local Ethnic Diversity (ELF)
Meatu, Tanzania: total local school funds Meatu, Tanzania: desks per primary
per pupil (2001 U.S. dollars) per year in school pupil in 2001 versus village ethno-
1997-2002 versus village ethnolinguistic linguistic fractionalization
fractionalization
FIGURE 2
between village diversity and local school funding there. For one out-
come—the proportion of water wells with normal flow—the coefficient
estimate on ELF is negative but statistically insignificant. However,
Table 3 (discussed below) provides more compelling evidence on the
water infrastructure in Meatu, Tanzania: the number of functioning
wells per household is not in fact any lower in diverse villages. Figure 2c
graphically presents the weak estimated relationship between ethnic di-
versity and school funding in Meatu, Tanzania, and Figure 2d presents
the relationship for desks; and these offer a sharp contrast to the nega-
tive relationships in Kenya.
TABLE 3
LOCAL PUBLIC FINANCE, COLLECTIVE ACTION, AND SOCIAL CAPITAL
OUTCOMES: TANZANIA3
Coefficient Estimate
Dependent Variable on LocalELFh
Panel A: Public Finance Outcomes, 2001—2 Village Council, Household Data

Annual total local expenditures on all public goods projects, 7.0
per household (U.S.S) (8.3)
Annual local expenditures on health and water well projects, 0.5
per household (U.S.S) (1.3)
Total annual local tax collection, per household 0.6
(2.7)
Average number of completed local public goods project, per year -0.33
(0.42)
Average household spending on local taxes and school expenses (U.S.S) 7.3
(11.0)
Panel B: Local Infrastructure
Wells with normal water flow, per household 0.002
(0.006)
Average road quality (scale 1-4) -0.0
(0-4)
Panel C: Community Group and Village Meeting Activity
Total community groups, per household (Source: Village Council Survey) 0.027
(0.017)
Household community group memberships, total (Source: Household Survey) -0.42
(0.69)
Proportion household survey respondents who are community group members -0.27**
(Source: Household Survey) (0.10)
Village meeting attendance, per household (Source: Village Council Survey) 0.75
(0.94)
Proportion households attending a village meeting, 2001-2 0.02
(Source: Household Survey) (0.07)
Panel D: Subjective Measures of Trust and Cooperation, 2001-2 Household Data

In general, can you trust people in other tribes? 0.20**
(0.09)
Spirit of cooperation across tribes in the village, proportion stating 0.26**
"above average" (0.10)
Village unity, proportion stating "above average" 0.13
(0.18)
Nation is more important than tribe to respondent -0.15
(O.H)
a
Huber robust standard errors in parentheses. Significantly different than zero at 90% (*), 95%
(**), 99% (***) confidence. Regression disturbance terms clustered at the ward level. The data is for 66
villages in Meatu, Tanzania. The socioeconomic controls are as in Table 2.
b
Using SUR across the twelve outcomes in Table 2 and Table 3 Panels A-B, the hypothesis that the
coefficient estimate on ELF is equal to zero is not rejected at traditional confidence levels
(F-statistic = 1.0, p-value = 0.46).
The key coefficient for our purposes is that on the ELF*KENYA
term, which captures how the relationship between ethnic diversity and
local public goods differs between Busia, Kenya, and Meatu, Tanzania.
We find that the coefficient estimate is negative in four of the five out-
comes we examine (with | t-statistic | > 1 for all four) and is negative
and statistically significant at 95 percent confidence for desks per pupil.
The seemingly unrelated regression (SUR) method combines informa-
tion across dependent variables; using this method we reject the hy-
pothesis that the coefficient estimate on ELF*KENYA is equal to zero
at over 95 percent confidence (F-statistic = 2.7, p-value = 0.02). In
other words, local ethnic diversity has a significantly more negative effect on
local public goods provision in Busia, Kenya, than in Meatu, Tanzania.
This is the main result of the article. Similarly, using SUR we reject the
hypothesis that the effect of local ethnic diversity in Kenya is zero at
over 95 percent confidence (F-statistic = 2.7, p-value = 0.02) but cannot
reject the hypothesis that ethnic diversity is unrelated to local public
goods outcomes in Tanzania (F-statistic = 1.0, p-value = 0.44).76
The SUR results are robust to aggregating the data up to the ward
level (for Meatu, Tanzania) and the zone level (for Busia, Kenya); the
sixty-six Tanzanian villages in our sample are located in nineteen wards
and the eighty-four Kenyan primary schools are in twenty-two zones
(results not shown). The main empirical results are also robust to the in-
clusion of a linear ethnic diversity measure—the proportion of the largest
ethnic group in the community—as an alternative diversity measure, and
robust to constraining the coefficient estimates on the socioeconomic
controls to be the same in the two countries (results not shown).
OTHER PUBLIC FINANCE AND SOCIAL OUTCOMES IN

MEATU, TANZANIA
The remaining results from the Tanzanian district relate to various ad-
ditional public finance and social capital measures. Since we do not
have comparable data from the Kenyan district, we are unable to em-
ploy the same identification strategy in estimating the impact of diver-
gent nation-building policies across the two countries.77 Nonetheless,
these relationships are interesting in their own right.
76
Ethnic diversity is unlikely to be proxying for higher local income inequality in Tanzanian villages,
since the correlation between diversity and inequality is small, negative, and not statistically significant
(regression not shown).
"We attempted to examine analogous issues in the Kenyan district, but data on registered commu-
nity group membership were only available for part of the study area. Restricting attention to regis-
tered community groups is also not ideal, since many groups are not registered in Kenya. The
relationship between local diversity and registered group membership in this limited sample is typi-
cally negative (not shown), but due to the data limitations mentioned above, we do not highlight the
Kenyan results.
358 WORLD POLITICS
Ethnic diversity is unrelated to a range of other local public finance

outcomes in Meatu, Tanzania. Total local expenditures per household
on public goods projects—perhaps the best estimate of overall village
council activity—is not significantly related to local ethnic diversity,
and the coefficient estimate is near zero (coefficient estimate U.S.$7.0,
standard error 8.3). The same holds for local expenditures on health
and water projects, total local tax collection, the number of completed
local public goods projects, and average spending on local taxes and
school expenses (Table 3, panel A). There is no evidence that the qual-
ity of local water well or road infrastructure is related to local ethnic di-
versity (panel B). The main results are robust to the inclusion of average
village per capita income and the village income Gini coefficient as ex-
planatory variables (results not shown). We cannot reject the joint hy-
pothesis that the coefficient on ELF'is equal to zero for all twelve local
public finance outcomes for Meatu presented in Table 2 and Table 3,
panels A-B (F-statistic = 1.0, p-value = 0.46).
The results on ethnic diversity and social capital in Meatu, Tanzania,
are mixed: there is no significant relationship between village ethnic di-
versity and the total number of community groups (usually self-help
groups) or with attendance at village meetings, and in fact the point es-
timates on ethnic diversity are positive though insignificant (Table 3,
panel C). There is, however, a strong negative relationship between
local ethnic diversity and the probability that a household survey re-
spondent was a member of a community group, echoing recent findings
from the United States.78 This membership effect is reasonably large:
the change from complete ethnic homogeneity to average levels of di-
versity in the district is associated with an 11 percent drop in group
membership.79 Yet individuals in diverse villages express significantly
higher levels of trust and cooperation across ethnic lines—perhaps due
to more frequent interactions, and thus closer social ties, with people
from other ethnic groups in diverse villages (panel D).
QUALITATIVE EVIDENCE FROM STRUCTURED INTERVIEWS
Evidence from structured interviews is consistent with the econometric

findings and sheds light on how nation-building policies have allowed
communities to overcome ethnic divisions in Tanzania.
The case of Matumbai Primary School in Kenya illustrates how low
78
Alesina and La Ferrara (fn. 8).
79
The correlation between social capital and ethnic diversity is similar in a nationally representative
sample of eighty-seven rural Tanzanian clusters, using data from the World Bank's 1995 Social Capi-
tal and Poverty Survey (SCPS) and 1993 Human Resource Development Survey (HRDS) (not shown).
interethnic cooperation in Kenya leads to reduced school investment.
Matumbai is one of the most ethnically diverse schools in the Kenyan
study district, with sizable ethnic Luhya,Teso, and Kalenjin communi-
ties. The headmaster of Matumbai stated in a June 2000 interview that
ethnic "rivalry over ownership" of the school and over which group "will
take control of the school" was the central challenge facing Matumbai.
Most parents have refused to participate in community fund-raisers
{barambees) or in school meetings in recent years due to a general lack
of trust across ethnic groups and the absence of a feeling of "ownership"
for the school. As a result, per pupil local school funding in the 1996
survey was one-third of average local funding in Busia and no class-
rooms had been constructed at the school. The result was that all
classes took place under a tree—which meant that school was canceled
when it rained. Many other headmasters report similar ethnic divisions
in western Kenyan schools, in the absence of an overarching national
identity to bind different groups together.
Miguel and Gugerty80 collected information from Kenyan primary
school records on the number of times that school committees imposed
sanctions on parents late with their school fees—for instance, embar-
rassing free riders by announcing the names of parents late with school
payments at village meetings or sending a local chief to their home to
pressure them to pay their fees. Miguel and Gugerty find that there are
significantly fewer sanctions imposed in ethnically diverse villages in
the Kenyan district, and the interview evidence mentioned above pro-
vides a possible explanation: in diverse Kenyan villages, where commu-
nity cohesion and trust are low, informal sanctions imposed across
ethnic lines are usually ineffective methods of pressuring individuals to
contribute to the public good.
In sharp contrast, ethnic divisions are reported to be minimal in the
Tanzanian study district: in the 2001 village council surveys, respond-
ents claimed that local ethnic relations were "good" in 97 percent of the
villages. Primary school committee members in Imalaseko village were
puzzled at the suggestion that ethnic divisions could play a role in local
school funding decisions. In fact, ethnicity played such a minor role in
Imalaseko that the committee even had difficulty assigning an ethnic
affiliation (Sukuma or Nyiramba?) to an absent committee member in
one case. A schoolteacher responded to a question about possible
ethnic divisions on the school committee in a November 2000 inter-
view by stating flatly: "This is Tanzania—we do not have that sort of
'"Miguel and Gugerty (fn. 6).

360 WORLD POLITICS
problem here." In an August 2002 interview, an official in Mwamishali
village explained that there was good cooperation across ethnic groups
because "we are all Tanzanians," and an elder in Mwambiti village re-
sponded similarly, suggesting good ethnic relations resulted from the
fact that "they [village residents] simply live as Tanzanians" ("Wanaishi
kama watanzania tu"; author's translation from Swahili).
The bottom line from the interviews is that while local politics in
Busia are characterized by ethnic "us versus them" appeals, such argu-
ments are considered illegitimate and downright "un-Tanzanian" in
Meatu. The elimination of ethnic appeals from acceptable political dis-
course may be the most important legacy of the Tanzanian nation-
building program. Tanzanian nation-building policies foster trust
across ethnic groups and a strong sense of identification with members
of other groups as fellow Tanzanians; these emotional bonds—together
with frequent village meetings and active local government institu-
tions—have allowed diverse Tanzanian communities to thrive where
diverse Kenyan communities fail.
VII. CONCLUSION
To summarize the main results, although western Kenya and western

Tanzania were similar along key dimensions in the 1960s, after inde-
pendence Tanzania adopted arguably the most serious nation-building
program in sub-Saharan Africa. And in forty years these regions appear
to have diverged: ethnic diversity leads to lower public goods funding in
western Kenya but is not associated with poor collective action out-
comes in western Tanzania.
Moving to the national level for further evidence, Tanzanian eco-
nomic growth rates were also substantially faster than Kenyan growth
rates during the 1990s, measures of governance and institutional qual-
ity consistently better, and national politics less violent.81 Although we
should not read too much into the national differences—which are the
product of many factors—these broad patterns are also consistent with
the claim that Tanzanian nation-building policies have indeed had a
beneficial long-run impact on political stability and economic develop-
ment. On a less formal level, visitors to Kenya and Tanzania are rou-
tinely struck by the different popular attitudes toward tribe and nation
in the two countries, and the far greater degree of attachment to na-
81
United Nations Development Program, 2002 Human Development Report: Deepening Democracy in
a Fragmented World (New York: Oxford University Press, 2002).
tional ideals, political leaders, and the Swahili language in Tanzania (al-
though the lingering separatist dispute with Zanzibar indicates that re-
gional divisions have not been entirely eliminated).
Yet there are also legitimate sources of concern regarding nation-
building policies: Tanzania may have successfully defused ethnic ten-
sions, but at what cost? First, the articulation and imposition of a single
national identity through coordinated public policies may have serious
negative costs for communities that do not fit neatly into the dominant
national vision, as the well-known examples of the United Kingdom82
and France83 illustrate. As a result, in many societies the fear remains
that the construction of a national identity will accelerate the erosion
of indigenous cultures and perhaps lead to a backlash by those who per-
ceive these policies as a threat to their way of life. Nation-building poli-
cies could also be employed by opportunistic ethnic majority leaders to
repress the legitimate political aspirations of minorities under the guise
of benign reform. Where ethnic divisions are pronounced, the process
of nation building may be slow, such that in the short term other solu-
tions—perhaps even the secession of regions dominated by dissident
minority groups—may lead to less conflict.
However, even if nation-building policies should not be promoted in every
ethnically diverse society, the Tanzanian case suggests that nation building
can succeed in an African context without jeopardizing indigenous cultures
and languages. Most vernacular languages—like the Sukuma language
in Meatu—continue to thrive in nonofficial contexts in Tanzania
decades after independence, coexisting with Swahili in homes and mar-
kets, just as the Tanzanian national identity coexists with ethnic identi-
ties rather than replacing them. In fact, one key to the success of the
Tanzanian reform program may be that the central government never
made efforts to stamp out vernacular languages or most indigenous cul-
tural practices, or to deny the very existence of particular ethnic groups.
Another reasonable concern about nation building is that, although
it binds people together within a society—reducing the likelihood of
domestic civil strife—it may provoke nationalistic sentiments that lead
to war with neighboring countries. However, once again, this fear has
not materialized in Tanzania. In fact, Tanzania has been an excellent
neighbor, accepting millions of refugees fleeing armed conflicts in the
region, and working consistently for negotiated settlements to regional
civil wars—most recently in Burundi. Internal stability and interna-
82
Linda Colley, Britons: Forging the Nation, 1707-1837 (New Haven: Yale University Press, 1992).
83
Eugen Weber, Peasants into Frenchmen: The Modernization ofRural France, 1870-1914 (Stanford,
Calif.: Stanford University Press, 1976).
362 WORLD POLITICS
tional peace have gone hand in hand for Tanzania, perhaps as a result of
the Pan-Africanist ideals at the heart of Nyerere's political philosophy.
Turing to policy implications, nation building will require a dramatic
restructuring of cultural, educational, and language policies in many
countries, and the centralized nature of this reform runs against current
"Washington Consensus" thinking about economic development,
which is suspicious of state-led development strategies. Nation building
in less developed countries is also likely to be opposed by powerful
politicians in the global North, concerned that increasingly nationalis-
tic regimes will promote anti-Northern views. Moreover, the benefits
of nation building may take decades to materialize.
Nonetheless, the results of this article suggest that the risks may be
worth taking, and that nation building should move onto government
policy agendas, especially in Africa. The articulation of new national
political identities and institutions has been under way in many African
countries since the democratization wave of the early 1990s, which re-
opened the public debate on the nature of the state in Africa. As a re-
sult, the coming decade may be a special window of opportunity,
similar to the postindependence period, for progressive African leaders
to adopt elements of the Tanzanian nation-building "model" as invest-
ments in long-run social stability and economic growth.

Paper Tribe or Nation Nation Building and Public Goods in Kenya Versus Tanzania

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Paper Tribe or Nation Nation Building and Public Goods in Kenya Versus Tanzania

Uploaded by

Copyright:

Available Formats

TRIBE OR NATION?

Nation Building and Public Goods in

T HE design of public policies that promote interethnic cooperation

World Politics 56 (April 2004), 327-62

III. A NATION-BUILDING CASE STUDY: KENYA VERSUS TANZANIA

GEOGRAPHIC AND HISTORICAL COMMONALITIES

examination of the countries. Both are populated mainly by small peasant

The two districts in this study—Busia, Kenya, and Meatu, Tanzania

cation was included as a standard curriculum subject in both primary

The empirical analysis utilizes household-level, school, and local gov-

DATA FROM BUSIA, KENYA

DATA FROM MEATU, TANZANIA

V. IDENTIFYING E T H N I C DIVERSITY EFFECTS

There are two steps in the econometric identification strategy. First, we

ETHNIC DIVERSITY IN KENYA

"Juma Volter Mwapachu, "Operation Planned Villages in Rural Tanzania: A Revolutionary

IDENTIFYING T H E IMPACT O F N A T I O N BUILDING

suggesting that our estimated effects may be lower bounds. A second

ECONOMETRIC ESTIMATION EQUATION

Y? = a* + X.' pk + TkELF. + uk (1)

Regression disturbance terms are assumed to be independent across

Yh = a\ + ai KENYA. + X. 'j8f + {X * KENYA.)'B*L

captures the impact of ethnic diversity on local outcomes in Tanzania

VI. EMPIRICAL RESULTS FROM WESTERN KENYA AND

Panel B: Datafor Tanzania

ETHNIC DIVERSITY AND LOCAL COLLECTIVE ACTION OUTCOMES

Annual Proportion H0:f5 = 0

Panel A: Public Finance Outcomes, 2001—2 Village Council, Household Data

Panel D: Subjective Measures of Trust and Cooperation, 2001-2 Household Data

OTHER PUBLIC FINANCE AND SOCIAL OUTCOMES IN

Ethnic diversity is unrelated to a range of other local public finance

QUALITATIVE EVIDENCE FROM STRUCTURED INTERVIEWS

Evidence from structured interviews is consistent with the econometric

'"Miguel and Gugerty (fn. 6).

To summarize the main results, although western Kenya and western

You might also like